Impact of rainfall and model resolution on sewer hydrodynamics
G. Brunia, J.A.E. ten Veldhuisa, F.H.L.R. Clemensa, b
a Water management Department, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Stevinweg 1, 2628CN, Delft, NL b Deltares, P.O. Box 177, 2600 MD Delft, The Netherlands
7th International Conference on Sewer Processes & NetworksWed 28 - Fri 30 August 2013
The Edge Conference Centre, Sheffield
Problem statementIMPROVEMENT OF RAINFALL ESTIMATE ACCURACY
ENHANCEMENT OF THE USE OF RADAR RAINFALL
X-BAND DUAL POLARIMETRIC RADARS TO INCREASE THE RESOLUTION AND ACCURACY
IMPROVEMENT OF THE DETAIL OF SEWER MODELS
ACCURATE INFORMATION OF SEWER CHARACTERISTICS AND PROCESSES
DETAILED LAND USE INFORMATION
?
• it serves 9 municipalities (~3’000 to ~43’000 inhabitants)
• 18.6 km long free-flow conduit • Sewer system detail: from~ 500 nodes to
3’800 nodes .
Casestudy
0 50 10025Kilometers
¯
Belgium
Germany
North Sea
The Netherlands
Eindoven area: Riool Zuid -> “Southern sewer system”
Luijksgestel
Bergeijk Westerhoven
Valkenswaard
Aalst
Waarde
Riethoven
END NODEPUMPING STATION
Veldhoven
Eindhoven-SE
0 6 123Kilometers
¯
Dataset-C-band radar data, KNMI
#*
#*
Luijksgestel
Bergeijk Westerhoven
Valkenswaard
Aalst
Waarde
Riethoven
RG-VESSEM
RG-BERGEIJK
RZ-END NODEPUMPING STATION
Veldhoven
0 1 2 3 40.5Kilometers
¯
Luijksgestel
Bergeijk Westerhoven
Valkenswaard
Aalst
Waarde
Riethoven
RZ-END NODEPUMPING STATION
Veldhoven
0 1 2 3 40.5Kilometers
¯
Ground measurements: 2 rain gauges Bergeijk and Vessem
Rainfall
Model lumping
Dataset- model specification • Software package: Infoworks
CS
• Sewer flow routing: dynamic wave approximation of the Saint-Venant equations
• Runoff estimation model: Fixed RC (impervious areas) and Horton for infiltration losses (pervious areas)• Runoff concentration: single linear resevoir (Desbordes)
Methodology1. RAD-
D 1. RAD-D: distributed sewer model with radar rainfall in input 2. RG-D: distributed sewer model with rain gauge rainfall in input 3. RAD-L: lumped sewer model with radar rainfall in input 4. RG-L: lumped sewer model with rain gauge rainfall in input
2. RG-D
4. RG-L3. RAD-L
Simulated scenarios
Rainfall selection-Event 1
Date Duration
Total volume range (mm)
Maximum intensity range (mm/h)
(dd-mm) (h) RG RAD RG RAD
12-7 2.50 9.9-2.8 9.2-0.7 37.2-
12 28.6-1.5Radar storm accumulation and maximum intensity
Storm evolution at Bergeijk and Vessem rain gauges vs overlapping
radar pixels
Rainfall selection- Event 2, 3 and 4
Event 4
Event 3
Event 2
Results-1 Water level results at Bergeijk
Luijksgestel
Bergeijk Westerhoven
Valkenswaard
Aalst
Waarde
Riethoven
END NODEPUMPING STATION
Veldhoven
Eindhoven-SE
0 6 123Kilometers
¯ Event 1 Event 2
Event 3 Event 4
MSE EV1 EV2 EV3 EV4RG-D vs RAD-
D 0.029 0.005 0.059 0.012RG-L vs RAD-L 0.001 0.00001 0.013 0.00006RAD-L vs RAD-
D 0.053 0.010 0.259 0.021RG-L vs RG-D 0.061 0.012 0.166 0.027
Results-2 Water level results at Westehoven-Event 2
Valkenswaard- Event 3
Luijksgestel
Bergeijk Westerhoven
Valkenswaard
Aalst
Waarde
Riethoven
END NODEPUMPING STATION
Veldhoven
Eindhoven-SE
0 6 123Kilometers
¯
12
3
4
Results 3- differences along the main conduit
1 2 3 4
1 2 3 4
Conclusions• The impact of model structure on water levels is
higher at locations close to rain gauges, i.e. when the rain gauge does accurately describe the storm evolution;
• The effect of rainfall resolution on model results becomes significant at locations far from the ground measurements: rain gauge fails to describe rainfall structure;• The bias found in all six scenario pairs increases in
the downstream direction, since the rain gauge is not representative of rainfall occurred at large distance (>4 km), and the lumping of larger catchments introduces higher error.
Thank you!
Main characteristics of Riool Zuid sewer system:
Luijksgestel
Bergeijk Westerhoven
Valkenswaard
Aalst
Waarde
Riethoven
END NODEPUMPING STATION
Veldhoven
Eindhoven-SE
0 6 123Kilometers
¯
Brandes spatial adjustment (BRA)*This spatial method was proposed by Brandes (1975). A correction factor is calculated at each rain gauge site. All the factors are then interpolated on the whole radar field. This method follows the Barnes objective analysis scheme based on a negative exponential weighting to produce the calibration field:
where di is the distance between the grid point and the gauge i. The parameter k controls the degree of smoothing in the Brandes method. It is assumed constant over the whole domain.The parameter k is computed as a function of the mean density of the network, given by the number of gauges divided by the total area. A simple inverse relation has been chosen:k = (2δ)−1
The factor 2 was adjusted to get an optimal k for the full network.The optimal k was estimated by trial and error based on the verification for the year 2006. The same relation between k and is used for the reduced networks.
*J.W. Wilson and E.A.Brandes, 1975 E. Goudenhoofdt and L. Delobbe, 2009
Evolution of bias – Bergeijk on event 1
16:1
9
16:4
0
17:0
2
17:2
4
17:4
5
18:0
7
18:2
8
18:5
0
19:1
2
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035 0
5
10
15
20
25
RGdistr - RADdistr
RGlump - RADlump
RADlump - RADdistr
RGlump - RGdistr
RAINFALL (RG)
BergeijkBIA
S= M
SE -σ
²e
Duration (h)
Rai
nfal
l Ber
geijk
RG
(mm
/h)